AI and Agentforce

Dreamforce 2026 Takeaways: Salesforce Is Rebuilding the Way Businesses Work with CRM

By Ankit Sharma9 min read
166 views

After three days of announcements, from September 15 to 17, one message from Salesforce stood out: the CRM interface we've grown accustomed to is no longer the center of the experience.

Salesforce Dreamforce 2026 wasn't simply about launching new features. It was about redefining how people, AI agents, data, and business workflows interact.

From AIforce and Agentforce to multi-model AI, new pricing editions with bundled AI credits, Salesforce Guardian, Agent Fabric, CCaaS, and a new generation of named AI agents, Salesforce is making a much bigger bet:

The future of CRM may involve less time inside Salesforce, and more intelligence working around it.

We covered the major announcements in our Dreamforce 2026 Keynote Recap. But announcements are only the starting point. The bigger question for Salesforce customers is:

What do these changes mean for your architecture, budget, governance, and roadmap?

Here are our Dreamforce 2026 takeaways at a glance, then the detail behind each.

Dreamforce 2026 Takeaways at a Glance

  1. The Salesforce UI is becoming optional. AIforce lets people work with CRM data from Slack, Claude or custom interfaces, still governed by Salesforce.
  2. Pricing now has two variables. Core ($195), Advanced ($395) and AIforce Max ($550) are priced per user per month and each includes Flex Credits, so Agentforce pricing depends on both seats and agent usage.
  3. Model choice is an architecture decision. Salesforce now pairs its own Koa CRM reasoning model with Claude, through Claudeforce and Salesforce in Claude.
  4. Agents now have names and jobs. Seven role-specific agents, from Casey (service) to Hunter (outbound sales), are ready to deploy, but each still needs an owner and rules.
  5. Governance can't wait. As agents multiply, every one of them needs an owner, clear permissions and escalation rules before it reaches production.

AIforce: Is the Salesforce Login Screen Becoming Optional?

The biggest announcement wasn't really a feature. It was a change in philosophy.

Timeline slide titled "AI is creating an interface revolution" tracing computing interfaces from the command line (1981) to the graphical user interface (1984), the web (1993), and mobile (2007), ending with AI today: an AIforce chat that answers a request for qualified leads with 242 leads and $200K in pipeline.
Salesforce frames AIforce as the next interface after the command line, the GUI, the web and mobile. Source: Salesforce, Dreamforce 2026 keynote.

AIforce, which Salesforce announced on September 15, is a new interface layer designed to let people work with CRM data and workflows from the tools they already use, including Slack, Claude, or custom interfaces, instead of requiring them to open Salesforce every time.

Under the hood, AIforce brings together several capabilities:

  • Claudeforce: Salesforce's deeper partnership with Anthropic.
  • Slackforce: CRM context and workflows inside Slack.
  • Agentforce Coworker: an AI teammate that works within Lightning.
  • Headless Toolkit: APIs, MCP servers, and developer tools for building custom experiences.

Salesforce co-founder Parker Harris captured the idea with a provocative question:

"Why should I ever log into Salesforce again? … Maybe you never will."

Parker Harris, Co-Founder, Salesforce

What does that mean for Salesforce adoption?

For years, Salesforce adoption has largely meant getting employees to spend more time in Salesforce. AIforce points toward a different model. Your CRM data and workflows remain governed by Salesforce, but users can interact with those capabilities wherever they already work.

That could mean less screen training, and more workflow and agent design. Instead of asking:

"How do we get our sales team to use Salesforce more?"

Organizations may increasingly ask:

"How do we bring Salesforce capabilities into the tools our sales team already uses?"

That is a significant shift in how Salesforce implementations should be designed.

Salesforce's New Editions: Core, Advanced and AIforce Max

AIforce wasn't the only major change. Salesforce also reorganized how it packages and prices the platform, with three new editions. For anyone researching Agentforce pricing, this is now the starting point: Agentforce is included in every edition.

Salesforce pricing slide comparing three editions: Core ($195), Advanced ($395), and AIforce Max ($550) per user per month. A checklist shows which features each tier includes, Flex Credits rise from 500K to 1M to 2.75M, and a callout offers a free upgrade from Agentforce 1 Edition.
Source: Salesforce, Dreamforce 2026 keynote.
EditionPriceFlex Credits included
Core$195 per user/month500K
Advanced$395 per user/month1M
AIforce Max$550 per user/month2.75M

Every edition includes Core Sales, Service and Industries, plus Agentforce. Even Core now adds Premier Success Plans, Slack Business with Slackbot, and Tableau Next. The higher editions build on that:

  • Advanced adds a Full Sandbox, Sales Programs, Casey (the help agent), Backup & Recover, Data Detect and Security Center.
  • AIforce Max adds Salesforce in Claude, an unmetered Agentforce Coworker, Momentum, Slack Enterprise+, Salesforce Spiff, Sales Planning, Service Rep Assistant, Workforce Management, Quality Management, Salesforce Maps and Revenue Intelligence.

Salesforce says customers on Agentforce 1 Edition can upgrade to the new Max edition at no extra cost.

Why Salesforce customers should care

Seats haven't gone away: every edition is still priced per user, per month. What's new is that each edition now includes a pool of Flex Credits, which AI agents draw down as they work. So your budget has two variables: how many users you license, and how much work their agents do. The financial conversation starts to shift from:

"How many Salesforce users do we have?"

to:

"How many users do we have, and how many workflows are we asking agents to perform?"

That makes AI planning both a technology and a financial exercise. Before your next Salesforce renewal or major Agentforce rollout, it is worth modeling:

  • Expected agent usage
  • Automated workflow volume
  • Whether your edition's included Flex Credits cover it
  • Cost per workflow
  • Business value generated
  • Human effort being augmented or replaced
  • Governance and monitoring costs

The goal isn't to automate everything. It's to understand where automation creates measurable value.

Claudeforce, Koa and Moirai: AI Model Choice Becomes an Architecture Decision

The Anthropic partnership, first announced as Claudeforce in August, took a major step forward at Dreamforce. Claude is now selectable as a reasoning model within Agentforce's Atlas Reasoning Engine.

Salesforce also introduced Salesforce in Claude, with 37 prebuilt sales skills covering use cases such as:

  • Meeting preparation
  • Pipeline review
  • Opportunity analysis

These capabilities operate using the permissions already established in Salesforce.

Illustration of a glass case on a base labeled "Salesforce Metadata Platform," holding a glowing network of connected cubes labeled Objects, Fields, Interfaces, Relationships, Rules, Permissions, Business Logic, Security, and Analytics, under the headline "Salesforce is powered by metadata. Metadata tells AI how your business works."
Your objects, rules and permissions live in Salesforce's metadata, which is why any model plugged in inherits them. Source: Salesforce, Dreamforce 2026 keynote.

Salesforce also unveiled two of its own models:

  • Koa: Salesforce's purpose-built CRM reasoning model.
  • Moirai: a forecasting model focused on demand and inventory planning.

Koa is built on NVIDIA Nemotron, and Salesforce's announcement states:

"No customer data was used to train the Koa reasoning model."

The bigger story

Salesforce isn't betting on one AI model. Instead, it is building an ecosystem in which different models can be used for different business problems. Claude may be appropriate for one workflow, a Salesforce model for another, and other models may eventually fit other use cases.

That means model selection is becoming an architecture decision, not simply a vendor decision. Salesforce architects and IT leaders will increasingly need to think about:

  • Which model handles which workflow?
  • What data can each model access?
  • How are permissions enforced?
  • How are costs monitored?
  • How can models be changed without rebuilding the entire process?

For a deeper comparison, see our Agentforce vs. Claudeforce post.

Marc Benioff: Trust, AI and the Future of Salesforce

Marc Benioff once again used the Dreamforce stage to address more than product announcements. Two themes stood out: trust and the future of the software industry.

On customer data, Benioff emphasized Salesforce's position that customer data remains protected and isn't used to train other models.

"Your data is your data.… You're not training any other model when you're using our products."

Marc Benioff, CEO, Salesforce

He also addressed the broader debate around whether AI-generated applications could eventually challenge traditional SaaS platforms. Benioff pushed back against the "SaaSpocalypse" narrative and pointed to Salesforce's recent business performance as evidence for the continued role of the platform.

Diagram titled "Enterprise intelligence runs on core systems" showing AI as probabilistic intelligence (planning, generation, reasoning) linked to a CRM cluster of governance, security, workflows, context, rules, and apps, which connects to a four-layer stack of interface, agents, apps and semantics, and data.
Salesforce's answer to the SaaSpocalypse: AI does the reasoning, but it runs on the CRM's governance, rules and data. Source: Salesforce, Dreamforce 2026 keynote.

Salesforce also announced $27 million in new education grants: $18 million for public school districts in San Francisco, Oakland, New York City, Chicago and Indianapolis, and $9 million for organizations supporting educators, with a focus on AI literacy.

The broader takeaway is that Salesforce isn't positioning AI as a side feature. It is positioning AI as the next stage of the platform itself.

Jensen Huang, Dario Amodei and the AI Debate

Dreamforce also brought together perspectives from some of the biggest names in AI.

Jensen Huang, CEO of NVIDIA, argued for moving quickly and cautioned against regulation slowing AI development, comparing the technology's arrival to the industrial revolution.

Dario Amodei, CEO of Anthropic, provided a more cautious perspective, emphasizing responsible development and industry-wide safety standards.

Patrick Stokes, President of Applications and Marketing at Salesforce, demonstrated how he built a Salesforce interface by talking to Claude, without traditional coding, and announced that the Claudeforce experience was entering open beta on AppExchange.

The contrasting perspectives reflected a broader question facing enterprise organizations:

How quickly should businesses adopt AI, and what controls should exist alongside that adoption?

For Salesforce customers, that question becomes increasingly practical as AI moves from experimentation into production workflows.

Meet the New Salesforce Agent Lineup

Salesforce also gave its AI agents something that makes them easier to understand: names and defined roles. Rather than asking every organization to build an agent from scratch, Salesforce introduced role-specific agents designed for specific business functions.

AgentPrimary role
CaseyCustomer service
FinCustomer service
PiperInbound engagement and pipeline generation
PaigeInternal IT and HR requests
CarterCommerce, product discovery and in-chat checkout
MarshallBack-office operations
HunterOutbound sales and long-horizon pipeline execution

Casey is already live on help.salesforce.com, where Salesforce credits it with resolving five million service conversations.

Fin comes from Intercom, which Salesforce agreed to acquire in June 2026.

Hunter is particularly interesting because it uses Salesforce's new long-horizon runtime, allowing it to pursue a sales objective across days or weeks rather than simply completing a single task. Salesforce plans general availability for Hunter in November 2026.

Easier deployment doesn't eliminate responsibility

Prebuilt agents lower the technical barrier to adoption. But they don't eliminate the need for governance. Before putting an agent into production, organizations still need to define:

  • What the agent can access
  • What actions it can take
  • What it cannot do
  • When human approval is required
  • How exceptions are handled
  • Who owns the agent
  • How success is measured

The easier agents become to deploy, the more important good governance becomes.

Agentic AI Is Expanding Across Salesforce

The AI story isn't limited to Agentforce. Salesforce demonstrated agent-driven capabilities across the broader Customer 360 platform.

Salesforce CloudWhat's changing
SalesHunter with integrated data vendors including ZoomInfo, Demandbase and Apollo
ServiceCCaaS launching in October
MarketingCampaign agent that can build and iterate campaigns
CommerceShopper agent for real-time personalization
Field ServiceVoice-to-text for hands-free form completion
IT & HRSelf-updating configuration management database
Revenue CloudRenewals agent that identifies opportunities and suggests upsells
TableauProactive insights
Industries500+ prebuilt industry-specific agents and actions

Individually, these announcements may not seem transformational. Together, they reveal something much bigger:

Agentic AI is moving from a bolt-on capability to an increasingly integrated part of Salesforce.

Salesforce CCaaS: Contact Center Consolidation Arrives

Salesforce confirmed that its Contact Center as a Service (CCaaS) offering will go live in October. The goal is straightforward: bring telephony, CRM, and contact center capabilities together instead of forcing service teams to work across multiple disconnected systems.

For organizations currently using a separate CCaaS provider, this creates an important evaluation point. But consolidation shouldn't be based on a product demo alone. Before making a decision, evaluate:

  • Language support
  • Communication channels
  • Telephony capabilities
  • Existing integrations
  • Data migration
  • Reporting
  • AI capabilities
  • Total operating impact

If your contact center contract is approaching renewal, Salesforce CCaaS deserves a place in the evaluation.

Salesforce Guardian and Agent Fabric: Who Is Watching the Agents?

As organizations deploy more agents, another problem emerges: how do you manage all of them? Salesforce addressed this with two major announcements.

Salesforce Guardian

Guardian is positioned as an evolution of Salesforce Shield, with capabilities focused on:

  • Agent identity
  • Risk detection
  • Data classification
  • Rogue-agent detection
  • Zero-data-retention enforcement

Agent Fabric

Agent Fabric evolves MuleSoft into a centralized registry for agents across environments including:

  • Salesforce
  • Azure
  • AWS
  • Google

It provides visibility into:

  • Data lineage
  • Latency
  • Errors
  • Agent activity
  • Cost

Rohan Kumar, President of Platform and Engineering at Salesforce, put the scale of the challenge into perspective:

"Hundreds of agents, thousands of agents – who knows? Millions of agents across all your teams, platforms and enterprises. How are you going to manage them?"

Rohan Kumar, President of Platform and Engineering, Salesforce

That's becoming an enterprise architecture question, not simply an AI question.

What Should Salesforce Leaders Do Now?

Dreamforce is full of announcements. The challenge is deciding which ones deserve a place on your roadmap. Here are six practical areas to evaluate.

Identify Workflows That Don't Need the Salesforce UI

Look for repetitive processes that could be surfaced through Slack, Claude, or another interface. Ask: does this workflow really require a user to open Salesforce?

Make AI Model Selection an Architecture Decision

Don't choose models simply because they are available. Evaluate them based on:

  • Use case
  • Data requirements
  • Security
  • Performance
  • Cost
  • Governance

Revisit Your Salesforce Budget

The new editions pair per-user pricing with included Flex Credits. Forecast both: your user counts, and the workflow and agent usage that will draw down those credits.

Establish Clear Agent Ownership

Every production agent should have an owner responsible for its:

  • Scope
  • Permissions
  • Data access
  • Escalation rules
  • Performance
  • Business outcomes

Put Governance in Place Before Agent Sprawl

If you're already moving beyond pilots, AI governance should be part of the implementation, not an afterthought. Define the rules before dozens of agents are operating across departments.

Evaluate CCaaS Before Your Next Renewal

If contact center consolidation is part of your roadmap, October's Salesforce CCaaS launch creates a natural opportunity to evaluate the platform against your current environment.

Dreamforce 2026 FAQ

What were the biggest Dreamforce 2026 takeaways?

Salesforce launched AIforce, an interface layer that lets people use Salesforce from Slack, Claude or custom interfaces; repackaged the platform into Core, Advanced and Max editions with bundled Flex Credits; introduced Koa, its own CRM reasoning model; and expanded Agentforce with seven named, role-specific AI agents.

How much do Salesforce's new editions cost?

Core costs $195, Advanced $395 and Max $550 per user per month. They include 500,000, 1 million and 2.75 million Flex Credits respectively.

How does Agentforce pricing work now?

Agentforce is included in every new edition. Each edition is still priced per user per month, and each includes a pool of Flex Credits for AI agent work, so budgets depend on both the number of users and how much work their agents do.

Can Agentforce 1 Edition customers upgrade to Max?

Yes. Salesforce says existing Agentforce 1 Edition customers can upgrade to the new Max edition at no extra cost.

What is AIforce?

AIforce is Salesforce's live interface layer, announced on September 15, 2026. It brings Salesforce data, workflows and business logic to wherever people and agents work, through four parts: Claudeforce, Slackforce, Agentforce Coworker and a Headless Toolkit.

What is Salesforce Koa?

Koa is Salesforce's first CRM reasoning model for Agentforce, built on NVIDIA Nemotron. Salesforce says no customer data was used to train it; its training data was built from synthetic scenarios.

What are Salesforce's new named AI agents?

Casey (customer service), Fin (customer experience), Piper (inbound pipeline), Paige (IT and HR), Carter (shopping), Marshall (supply chain) and Hunter (outbound sales). Hunter is in pilot, with general availability planned for November 2026; the other six are generally available.

From Dreamforce Announcements to Your Salesforce Roadmap

Every Dreamforce leaves organizations with the same challenge:

A long list of new capabilities, and a much shorter list of people who know how to implement them without disrupting what already works.

That's where strategy and implementation experience matter. At Brazen Group, we help organizations turn Salesforce technology into practical Revenue Operations solutions. Our Salesforce expertise spans:

  • Sales Cloud
  • Service Cloud
  • Marketing Cloud
  • Revenue Cloud
  • Experience Cloud
  • Agentforce
  • Salesforce integrations
  • Revenue Operations strategy

And as AI becomes increasingly embedded in Salesforce, the questions we're helping organizations think through are changing:

  • Which workflows are ready to become agent-driven?
  • How should multiple AI models fit into the architecture?
  • How do we protect customer data while expanding AI access?
  • How should we plan for Salesforce's new editions and Flex Credits?
  • What governance needs to exist before agents move into production?

The answer isn't to implement every new capability Salesforce announces. The opportunity is to identify where AI can improve your specific business processes, and build the architecture, governance, and operating model to support it.

Dreamforce 2026 showed us where Salesforce is heading.

Now the question is: what should that future look like for your organization?

If you're evaluating Agentforce, AIforce, Salesforce's new AI architecture, or your broader Salesforce roadmap, Brazen Group can help you turn the announcements into an actionable plan.

Talk to Us